The brain that learns YOUR brain.
Most ed-tech sticks a chatbot on top. We built an AI that understands which 3 topics you should drill tonight, predicts your future-score range, and builds the plan to get there.
Built on your data, not the internet's
Every attempt updates 6 models. Every Sunday your study plan refreshes. Every night your weak-topic shortlist gets rebuilt. You see the rationale on every recommendation.
Six pillars
Each pillar is a documented feature with an ID, owner, model registry entry, and weekly calibration job.
It knows your strengths
Per-topic Bayesian skill model updated after every MCQ. Not just accuracy — it understands difficulty, recency, and uncertainty.
It tells you what to fix
Surfaces the 5 topics where one more hour of practice will maximally move your projected score. Updated every night.
It picks the right question type
You're 78% on Mechanics conceptual but 41% on Mechanics numericals. The AI knows. Tonight's 15 MCQs are exactly those.
It predicts your future score
Calibrated distribution (10th / 50th / 90th percentile). Always a range. Recalibrated weekly against the previous cohort's actuals.
It builds your study plan
Given exam date, hours-per-week, current mastery, target score — generates a week-by-week plan. Re-plans every Sunday based on actuals.
It coaches you 24/7
AI tutor that knows YOUR mistakes. Cited responses. Refuses non-exam questions. Never gives a single-number prediction.
Future-score · as a range
"If you stay on this pace, your NEET projected band is 612 to 668 (50th: 642). To cross 680 by exam day you need ~6 more hrs/week on Biology Numericals."
Single-number predictions are forbidden by policy. Bands are always shown with the model version + last calibration date.
Every MCQ updates six things
You tap an answer; six pipelines fire. None of them block your next MCQ — all are async or sub-500ms.
A coach who has read your notebook
Six guardrails
AI that decides your future has to be auditable, opt-outable, and humble about uncertainty.
Explainability
"Why this?" on every recommendation shows the 3 inputs that drove it.
Opt-out
Disable any AI feature in Settings. App still works (falls back to defaults).
No training without consent
Default OFF for "help improve Prep with my anonymised data".
Safety filter
All LLM outputs pass a safety classifier before being shown.
Range-only predictions
Single-number score predictions are forbidden. Always 10/50/90th.
Human override
Teacher / parent / you can pin a topic to override AI recommendation. Pinning logged.
docs/features/41_ai_prep_intelligence.mdLet the AI handle "what to study next"
You bring the discipline. The AI brings the strategy.